The RUBI Project: A Progress Report.

Page created by Johnnie Ortiz
 
CONTINUE READING
The RUBI Project: A Progress Report.
The RUBI Project: A Progress Report.

                                              ∗
                   J. R. Movellan                 1,2,3
                                                          , F. Tanaka1,2,3,5 , I. R. Fasel1,2,3 , C. Taylor1,4 , P. Ruvolo1,4 ,
                                                                    M. Eckhardt1,2,3
                                                  Machine Perception Laboratory, UCSD1
                                                 Institute for Neural Computation, UCSD2
                                                                CalIT2, UCSD3
                                           Computer Science and Engineering Department, UCSD4
                                                              Sony Corporation5

ABSTRACT                                                                             years ago we started a project, named RUBI, to pursue three
The goal of the RUBI project is to accelerate progress in                            key insights: (1) Progress can be accelerated by developing
the development of social robots by addressing the problem                           robotic systems immersed in the environment in which they
at multiple levels, including the development of new scien-                          need to operate. This allows feedback to occur on a daily ba-
tific methods, formal approaches, hardware, software, and                            sis, facilitating the rapid discovery of problems that need to
scientific agenda. The project is based on the idea that                             be solved. (2) Social Robotics needs to emerge as a coherent
progress will go hand-in-hand with the emergence of a new                            program that cuts across scientific disciplines. For example,
scientific discipline that focuses on understanding the orga-                        decoupling the study of human robot-interaction from com-
nization of adaptive behavior in real-time within the envi-                          putational analysis and hardware development may be detri-
ronments in which organisms operate. As such, the RUBI                               mental in the long run. (3) Progress requires a mathematical
project emphasizes the process of design by immersion, i.e.,                         framework for the analysis of real time social interaction.
embedding scientists, engineers and robots in everyday life                             Following Solomon Asch’s views on the emergence of new
environments so as to have these environments shape the                              scientific disciplines [2], we avoided rushing into controlled
hardware, software, and scientific questions as early as pos-                        conditions and laboratory experiments. Instead our focus
sible in the development process. The focus has been on                              was on observation and identification of scientific questions
social robots that interact with 18 to 24 month old toddlers                         and technological challenges. As such we emphasized the
as part of their daily activities at the Early Childhood Ed-                         process of design by immersion. The goal was to have real
ucation Center at the University of California, San Diego.                           life environments shape the hardware, software, and scien-
In this document we present an overall assessment of the                             tific goals as early as possible in the development process.
lessons and progress through year two of the project.                                We were also particularly inspired by David Marr’s philoso-
                                                                                     phy of science [28] which emphasizes the importance of com-
                                                                                     putational analysis, i.e., understanding the computational
Keywords                                                                             nature of the problems that organisms solve when operating
Design by Immersion, Field Studies, Social Robots, Archi-                            in their daily environments.
tectures for Social Interaction                                                         We decided to focus the project on the development of
                                                                                     robot technologies that can interact with toddlers (18-24
1.     PHILOSOPHY OF THE RUBI PROJECT                                                months of age). Children at this age were chosen because
                                                                                     they have few preconceived notions of robots and because
   The development of social robots brings a wealth of sci-                          we believed they would help focus our work on problems
entific questions and technological challenges that are only                         we deemed particularly important, e.g., timing, non-verbal
starting to be addressed in a coordinated manner [37, 11,                            communication, and the development of affective and social
50, 10, 18, 21, 35, 40, 24, 36, 34]. As such, progress is likely                     bonding behaviors.
to co-occur with the emergence of a Kuhnian-style “Scien-                               In its two years of life the RUBI project has generated a
tific Revolution” [25], i.e., a shift in focus and methods to-                       wealth of scientific articles spanning behavioral analysis of
wards the computational analysis of real-time social interac-                        human-robot interaction [44, 45, 46, 23, 15], new machine
tion in everyday environments. With this idea in mind, two                           perception primitives [6, 43, 27, 5, 4, 42], new machine learn-
∗Please send correspondance to movellan@mplab.ucsd.edu                               ing algorithms [12, 31], and documentation of the process of
                                                                                     design by immersion [32, 19, 30]. The specifics of the dis-
                                                                                     coveries emerging from the project can be found in these
                                                                                     articles. Here we focus on a general overview of the progress
Permission to make digital or hard copies of all or part of this work for            and experiences accumulated during the first two years of
personal or classroom use is granted without fee provided that copies are            the project. First we describe the origins and time-line of
not made or distributed for profit or commercial advantage and that copies           the RUBI project. Second we summarize the progress and
bear this notice and the full citation on the first page. To copy otherwise, to      research activities for Years 1 and 2. Finally we provide an
republish, to post on servers or to redistribute to lists, requires prior specific   overview of our general experience in the project and the
permission and/or a fee.
HRI’07, March 10–12, 2007, Arlington, Virginia, USA.
                                                                                     difficulties we found implementing the design by immersion
Copyright 2007 ACM 978-1-59593-617-2/07/0003 ...$5.00.                               approach.
The RUBI Project: A Progress Report.
2.    ORIGINS AND TIMELINE                                       the head”. Time series analysis revealed that haptic behav-
   The conceptual seeds of the RUBI project date back to         iors were surprisingly effective predictors of the perceived
an NSF-ITR project to develop machine perception primi-          quality of interaction. A linear combination of the output
tives, e.g., expression recognition, for autonomous tutoring     of low-pass filtered touch sensors could predict the frame by
systems [13]. We realized that while such perception primi-      frame quality of interaction, as assessed by humans, very
tives can be developed using current technology, the problem     well. While the interpretation of this result is still unclear it
of how to connect perception and action to produce fluid so-     helped raise our awareness about the special role that touch
cial behavior in real time is far less understood. The RUBI      and haptic behaviors play in social interaction.
project started in September 2004 [32] with an aim towards          We also analyzed the results of an experiment conducted
addressing some of these issues. The project is still evolv-     in Year 1. The goal of the experiment was to evaluate two
ing and is currently operating under the auspices of the UC      different dancing algorithms for social robots [44]. The study
Discovery Program. Below we describe the main research           lasted 6 field sessions at ECEC, 30 minutes each. For three
activities and results obtained during Years 1 and 2.            randomly selected sessions, QRIO was controlled by a chore-
                                                                 ographed dance program. For the other three sessions it was
                                                                 controlled by an optic-flow based dancing algorithm [47].
3.    SUMMARY OF YEAR 1: DEVELOPMENT                             The study showed that a simple algorithm that responds to
      AND DATA GATHERING                                         the motion of people was as compelling as a labor-intensive
   The field studies in the RUBI project are being conducted     choreographed dance program. Most importantly the study
at the UCSD Early Childhood Education Center (ECEC).             taught us that it is possible to run experiments, not just
The first 6 months of the project were spent volunteering at     observational studies, in the relatively uncontrolled condi-
the ECEC. This time was important for bonding with the           tions of daily life. We obtained replicable results in periods
children, teachers, and parents, and for developing a sense      of time that were shorter than those typically required for
of the problems that were likely to be encountered. It also      laboratory experiments. In fact we feel strongly that the
helped shape the general philosophy of the RUBI project,         experiment would have been very difficult to conduct in a
as expressed in the previous sections.                           laboratory setting detached from the daily conditions and
   The next 6 months were dedicated to conducting field ses-     activities of the children.
sions with two robots: RUBI, and QRIO. RUBI is a robot
platform that is being designed from the ground up by im-
mersion in the classroom. QRIO is a state of the art hu-
manoid robot prototype developed by Sony Corporation for
                                                                        RUBI's             Indoor
research purposes [1, 26, 20]. During these 6 months we                                                           Outdoor
                                                                        Room              Playroom
conducted in one room of ECEC a total of 60 field sessions                                                       Playground
(See Figure 1-Top) . All the sessions were taped with two
synchronized cameras for further analysis.
                                                                           Control
4.    SUMMARY OF YEAR 2                                                    Room
   In Year 2 the focus was on analysis of the 60 videotaped
field sessions and on redesigning the software for the RUBI
robot prototype based on the lessons we learned as part of
Year 1.

4.1   Analysis of the Field Studies
   Developing efficient methods to analyze the field sessions
in a manner that suited our goals was not a trivial task. Over
time we found two methods particularly useful: (1) The
continuous audience response methods used in marketing
research [38, 17], and (2) Frame-by frame labeling for the
Presence/Absence of target behaviors.
   Regarding the continuous audience response methods, we
developed software that allowed observers to operate a dial      Figure 1: Top: Layout of Room 1 at ECEC. There are three
in real time while viewing the video-taped sessions. The         playing spaces. Children are free to move back and forth between
position of this dial indicated the observer’s impression of     spaces thus providing information about their preferences. Bot-
the quality of interaction seen in the video. 30 times per       tom: Judges used a dial to continuously evaluate the quality
second the program recorded the position of the dial and         of the interaction between children and robots. Judges can si-
the video frame that the observers were viewing at that          multaneously see two synchronized movies taken by two separate
moment. Overlaid on the video, the observers could see           cameras. They can also see the recent history of their evaluation
a curve displaying their recent evaluation history (See Fig-     which is superimposed on the movie as a red graph.
ure 1-Bottom). We found that in spite of the abstract na-
ture of the dimension being coded (quality of interaction)
inter-observer reliability was quite high ( average Pearson
Correlation between 5 independent observers was 0.79) [44].      4.2    Mathematical Formalization
   We also coded, frame by frame, the presence or absence of       A critical challenge in Social Robotics will be the devel-
a variety of objective behaviors, e.g., “QRIO was touched on     opment of a mathematical framework for formalizing how
The RUBI Project: A Progress Report.
to connect perception and action in the context of real time
social interaction. As part of the RUBI project we took
significant steps towards such a goal. The framework we
are pursuing is based on the Theory of Stochastic Optimal
Control, an area of mathematics and engineering that deals
with the control of probabilistic processes [7]. It focuses on
the problem of finding control laws, i.e., moment-to-moment
mappings between perceptions, internal states, and actions,
to achieve long-term goals. The theory of stochastic op-
timal control was designed to solve many of the problems
that have been elusive to traditional cognitive and symbolic
AI approaches, particularly in regard to social interaction:
(1) the importance of timing; (2) the fact that social in-
teraction is a continuous “dance” rather than a turn-taking
process; (3) the need to act intelligently in the presence of
constantly changing uncertain information. In the everyday
world the human brain faces many control problems when
sending motor commands to the body’s actuators. Riding
a bicycle, using a computer mouse, shooting baskets, and
playing music are all control problems. We and others be-
lieve that real time social interaction is also, in essence, a
control problem. The parameters of the social interaction
problem are different from the parameters of the physical
interaction problem but the mathematical structure of these
two processes is identical [33, 51].
   We developed an example of how stochastic optimal con-
trol can be used to formalize real time social behavior. In
particular we focused on the problem faced by two month old
infants, given their limited perceptual capabilities, of detect-
ing the presence of responsive human caregivers [31]. The
idea behind the approach was that humans can be identified         Figure 2: Three real time control problems: person control-
by the temporal characteristics of their response to others,       ling a computer mouse, infant playing smile games with Mom,
a source of information commonly known as “social contin-          RoboVie-I playing with a person.
gency” in infancy learning literature [3, 8, 48, 49]. From this
point of view the problem faced by infants is that of detect-
ing the “social contingency signature” hidden in the stream        The architecture is designed to handle timing, uncertainty
of activity continuously received by their sensors. Once the       and learning in a network of goal-oriented message-passing
problem was formalized this way, an optimal controller was         nodes. The ultimate goal in RUBIOS is to have the pro-
developed that scheduled simple vocalizations, moment to           grammer focus on the goals the robot and let probability
moment, so as to detect social contingency as quickly and          theory and machine learning take care of the details of how
as accurately as possible. The optimal controller exhibited        to best achieve those goals. Each node in a RUBIOS robot
some interesting properties: (1) It modeled well the tem-          implements a parameterized “control law”, i.e., a function
poral dynamics of vocalizations found in social contingency        that couples the history of past sensor information and inter-
experiments with infants [39, 33]. (2) Turn taking behav-          nal representations to current actuator outputs. In addition
iors emerged in the controller as an optimal strategy. Most        each node has a learning processes whose role is to change
importantly these “turns” were not fixed apriori. Its length,      the node’s parameters to maximize the long-term pay-off
for example, changed dynamically based on the incoming             accumulated by that node.
sensory information. The algorithm was implemented in a               Nodes can affect each other by offering rewards that vary
humanoid robot and used as a primitive that allowed it to          as a function of dimensions such as time or similarity be-
learn on its own how to detect human faces [12].                   tween the desired goal and alternative goals. For example, a
                                                                   node may offer a reward for positioning a perceived face as
                                                                   close as possible to a desired location on the retina. In the
5.    SOFTWARE DEVELOPMENT                                         default implementation rewards vary as a function of time
   The design by immersion process was particularly influ-         and space in an exponential/quadratic manner:
ential in two aspects of software development: (1) The de-
                                                                                                     (x−ξ)2      t
velopment of a new software architecture for social robotics.                                    −
                                                                                    ρ(t, x) = ρ0 e     β      e− α           (1)
(2) The development of machine learning and machine per-
ception primitives robust enough to operate in field condi-        where ρ0 is the peak reward, α the time scale, β the space
tions.                                                             scale, and ξ the desired position. Here “space” refers to
                                                                   an abstract similarity space between states. Within the
5.1    Software Architecture: RUBIOS                               RUBIOS framework reflexive processes correspond to offers
  We developed the first version of RUBIOS, a software ar-         with high peak value and short time constants. Moods on
chitecture inspired on the ideas of stochastic optimal con-        the other hand emerge as a result of offers with small peak
trol to address issues encountered in the field during Year 1.     values but very long time constants.
The RUBI Project: A Progress Report.
The goal of each RUBIOS node is the long term maximiza-          pened to discriminate the presence of people very well.
tion of rewards. To do so each node has a simple control law,
which typically involves a greedy controller, and a learning        5.4   Hardware Development
process that is in charge of optimizing the default controller.
Nodes can pass information to each other via multiple chan-            RUBI’s robot design was inspired by Hiroshi Ishiguro’s
nels, which are typically optimized for the type of message         RoboVie-I humanoid [22, 29]. However, we found that the
being passed. For example, images are passed via mem-               RoboVie-I design was frightening to children under 4 years
ory mapping, while low-bandwidth messages are typically             of age and thus we systematically changed RUBI’s appear-
passed via sockets. The current version of RUBIOS consists          ance until children found it non-threatening. Some of the
of a set of classes specialized on different aspects of robot op-   modifications included shortening the body, making it more
erations: Inter-node communications, Interface with human           plump, including facial expressions, clothes, a touch-screen
operator, Node monitoring, Servo Control, Game control,             and hair. The current RUBI prototype is a three-foot tall,
and Vision.                                                         pleasantly plump robot with a head, two arms, and a touch
   While our experience with the current version of RUBIOS          screen (See Figure 4). The exterior appearance of RUBI
is still very limited, it holds promise in terms of the ease with   has been quite successful. In general the children found it
which different programmers can seamlessly add nodes that           non-threatening and by the end of the 13th session they ex-
integrate with the overall robot behavior. For example, en-         hibited a variety of social behaviors towards her including
ergy saving was implemented by adding a constant request            pointing, hugging, imitation, and social referencing. In Year
for all the servos to move to their resting point, yet hav-         2 we completely redesigned RUBI’s hardware while keeping
ing that request have a very small peak value and a very            her external appearance relatively unchanged.
long time constant. Adaptive reflexes were also easily im-             The latest version of RUBI is constructed from two Apple
plemented by adding nodes that produce requests with very           PowerMac desktop computers. Each unit has two 2.5 GHz
large peak values but very short time scales.                       dual-core PowerPC 970MP processors with 8GB of error
                                                                    correcting (ECC) RAM. One machine has an 802.11g wire-
5.2    Perceptual Primitives                                        less card used to control the robot from a nearby wireless
   For the past 15 years our laboratory has focused on the          laptop during sessions. Currently RUBI’s full software suite
development of perceptual primitives for social interaction         runs comfortably on a Mac Mini with a single 1.8 GHz Intel
(e.g., face detection, expression recognition, audio visual         Core Duo processor and 2GB of RAM, even while using two
speech recognition). Over the years these systems have been         cameras with face, eye, and smile detection and color track-
refined and are now operating in or near real time. RUBI’s          ing running on both. Thus the two quad G5 systems give
software include the latest versions of face detection, face        RUBI ample room to grow, and is essentially a mobile 8-node
tracking and video-based emotion recognition developed at           cluster. RUBI is capable of running the learning primitives
the laboratory. During Year 1 we found that while our sys-          mentioned in the previous section [16], which require large
tems worked well in controlled laboratory conditions, they          amounts of memory and 64-bit processing. RUBI’s sensors
did not work reliably enough in the relatively uncontrolled         and actuators are organized as a distributed network of con-
conditions of the classroom. In Year 2 we focused on the            trol and sensing nodes, all connected to the main computer
development of a robust face finder and smile detector that         via multiple USB ports. This significantly increases com-
could provide social robots with reliable estimates of this         munication bandwidth and avoids the bottleneck associated
important social behavior. The system was trained with a            with having a master controller in charge of all the servos
dataset of 70,000 images collected from the Web, containing         and actuators.
a very wide variety of imaging conditions, races, physical ap-         By far, RUBI’s arms has proven the most challenging,
pearances, etc. The new smile detector has a performance            frustrating, and elusive hardware design problem encoun-
level of 96.8 % on the dataset and can run in real time at          tered in the project. The difficulty lies on the need for ac-
standard video rates. The system is reliable enough to be           tuators that can handle the forces applied by children when
used in a wide variety of applications in real-life situations      interacting with RUBI, yet compliant and small enough to
and shall be one of the perceptual primitives in the new            be safe. Most importantly all of this needs to be done within
RUBI prototype.                                                     a very tight budget. In Year 2 we developed a streamlined
                                                                    5 degrees of freedom prototype that used high end robotic
5.3     Learning Primitives                                         RC servos. Unfortunately we ran into two problems: (1)
                                                                    The servos proved to be too noisy having a significant effect
   We developed a new approach for robots to learn to dis-
                                                                    on the interaction with the children. (2) While they worked
cover, in an autonomous manner, the visual appearance of
                                                                    well in our laboratory tests, they did not survive the rigors
objects in the world they operate [16]. In particular we con-
                                                                    of interaction with the children during the field studies. We
ducted a promising experiment showing how a social robot
                                                                    are currently working on our third arm design. Critical to
can learn on its own to detect faces. After less than 6 min-
                                                                    the new design is the issue of compliance control.
utes of interaction with the world, the robot’s visual system
was capable of detecting the presence of people in novel im-
ages with high accuracy (over 90 % correct).
   During the 6 minutes of exposure to the world, the baby
robot was never told whether or not people were present in
the images, or whether people were of any particular rele-          6.    LESSONS AND CONCLUSIONS
vance at all. It discovered that the most consistent visual           The RUBI project was conceived as an ambitious experi-
explanation for the cause of the observed sensory-motor con-        ment aimed at accelerating the development of robots that
tingencies was a combination of feature detectors that hap-         interact with people in everyday conditions [41, 14, 9, 22,
the conditions of daily life. (8) Identify the importance of
                                                                  studying how organisms organize behavior at multiple time
                                                                  scales: from reflexes to moods, emotions, and developmental
                                                                  processes.
                                                                     We also identified lessons and limitations of the design by
                                                                  immersion approach as originally conceived. We were naive
                                                                  in the idea that we could just immerse ourselves in field con-
                                                                  ditions on a daily basis and make incremental changes until
                                                                  we design a “dream social robot”. In practice qualitative
 Figure 3: A typical view from RUBI’s wide angle cameras          changes are needed in the hardware and software architec-
                                                                  ture that can take months if not years away from the field.
                                                                  We also found that some of the intuitions initially drawn
                                                                  from the field sessions turned out to be misguided, per-
                                                                  haps setting us back in time. For example, initially we felt
                                                                  that self-locomotion was a critical component for progress.
                                                                  We invested time and effort to develop a new version of
                                                                  RUBI that could move autonomously about the room, only
                                                                  to discover that self-locomotion was perhaps distracting us
                                                                  away from the main focus of the project– social interaction.
                                                                  We under-estimated the difficulties of mechanical and sen-
                                                                  sor technology issues faced by social robots. For example,
                                                                  we have not managed to develop a robot arm that operates
                                                                  robustly in field conditions. We also underestimated the role
                                                                  that controlled laboratory experiments may play for testing
                                                                  hypotheses of interest. While the field conditions proved
                                                                  useful for generating hypotheses it is difficult to eliminate
                                                                  alternative explanations and find conclusive evidence using
                                                                  field studies alone. A combination of field studies and tar-
                                                                  geted laboratory experiments may be a better strategy for
                                                                  progress.
Figure 4: RUBI teaching materials targeted by the California         Overall, we believe the RUBI project is turning out to be
Results Developmental Profile from the California Department of   an exciting and useful experiment that illustrates how an
Education.                                                        immersive paradigm can help make significant progress in
                                                                  the emerging field of social robotics.

29]. The core principle of the project is the idea of design      7. ACKNOWLEDGMENTS
by immersion, i.e., the immersion of scientists, engineers and      This study is ongoing at Room-1 of UCSD’s Early Child-
robots in the conditions of daily life as early as possible in    hood Education Center. We thank the director of ECEC
the development process.                                          Kathryn Owen, the head teacher of Room-1 Lydia Morrison,
   We believe this is important for the development of hard-      and the parents and children of Room-1 for their support.
ware, software, the discovery of the technological and scien-     The RUBI project is funded by a UC Discovery Grant.
tific challenges whose solution may maximize progress, and
the development of a theoretical framework for robot con-         8. REFERENCES
trol. Rather than focusing on solving complex problems
in the controlled conditions of laboratory environments we         [1] http://www.sony.net/SonyInfo/QRIO.
focused on solving simpler problems in the uncontrolled con-       [2] S. E. Asch. Social Psychology. Oxford University
ditions of daily life.                                                 Press, New York, 1987 (Original work published in
   After two years we are as convinced as ever that the                1952).
design by immersion philosophy is sound and healthy. It            [3] L. R. Bahrick and J. S. Watson. Detection of
helped us: (1) Design machine perception primitives (e.g.,             intermodal proprioceptive-visual contingency as a
smile detection) that work well in field conditions, not just          potential basis of self-perception in infancy.
on the standard face datasets. (2) Develop machine learning            Developmental Psychology, 21:963–973, 1985.
primitives that can operate continuously in an unsupervised        [4] M. Bartlett, G. Littelwort, M. G. Frank, C. Lainscsek,
manner. (3) Formulate a mathematical approach to real                  I. Fasel, and J. Movellan. Recognizing facial
time social interaction to handle the timing and uncertainty           expression: Machine learning and application to
conditions of daily life. (4) Develop RUBIOS, a prototype              spontaneous behavior. In IEEE International
software architecture for social robots. (5) Establish that            Conference on Computer Vision and Pattern
long term socialization and bonding can develop between                Recognition, pages 568–573, Osaka, Japan, 2005.
humans and robots, at least when the robot is partially            [5] M. Bartlett, G. Littlewort, L. C., F. I., F. M.G., and
controlled by a human being. (6) Identify the particularly             M. J.R. Fully automatic facial action recognition in
important role of touch and haptic behaviors in the devel-             spontaneous behavior. In 7th International Conference
opment of this bonding process. (7) Develop methods for                on Automatic Face and Gesture Recognition, pages
evaluating social robot algorithms in an efficient manner in           223–228, 2006.
[6] M. S. Bartlett, G. Littlewort, C. Lainscsek, I. Fasel,            R. Nakatsu. Development and evaluation of an
     and J. Movellan. Recognition of facial actions in                 interactive humanoid robot ”robovie.
     spontaneous expressions,. Journal of Multimedia, in        [23]   T. Kanda, N. Miralles, M. Shiomi, T. Miyashita,
     press.                                                            I. Fasel, J. R. Movellan, and H. Ishiguro. Face-to-face
 [7] D. Bertsekas and S. Shreve. Stochastic Optimal                    interactive humanoid robot. IEEE 2004 International
     Control. Athena Scientific, 1996.                                 Conference on Robotics and Automation, 2004.
 [8] A. E. Bigelow. Infant’s sensitivity to imperfect           [24]   H. Kozima, C. Nakagawa, and Y. Yasuda. Interactive
     contingency in social interaction. In P. Rochat, editor,          Robots for Communication-care: A Case-study in
     Early social cognition: understanding others in the               Autism Therapy. In Proceedings of the 2005 IEEE
     first months of life, pages 241–256. LEA, New York,               International Workshop on Robot and Human
     1999.                                                             Interactive Communication, pages 341–346, 2005.
 [9] C. Breazeal. Designing Sociable Robots. MIT Press,         [25]   T. S. Kuhn. The Structure of Scientific Revolutions.
     Cambridge, MA, 2002.                                              University of Chicago Press, Chicago, 1996.
[10] C. L. Breazeal. Designing Sociable Robots. The MIT         [26]   Y. Kuroki, T. Fukushima, K. Nagasaka, T. Moridaira,
     Press, 2002.                                                      T. T. Doi, and J. Yamaguchi. A Small Biped
[11] R. A. Brooks, C. Breazeal, M. Marjanovic,                         Entertainment Robot Exploring Human-Robot
     B. Scassellati, and M. M. Williamson. The Cog                     Interactive Applications. In Proceedings of the 2003
     Project: Building a Humanoid Robot. Lecture Notes                 IEEE International Workshop on Robot and Human
     in Artificial Intelligence, 1562:52–87, 1999.                     Interactive Communication, pages 303–308, 2003.
[12] N. J. Butko, I. Fasel, and J. R. Movellan. Learning        [27]   G. Littlewort, M. Bartlett, I. Fasel, J. Susskind, and
     about humans during the first 6 minutes of life. In               J. R. Movellan. An automatic system for measuring
     International Conference on Development and                       facial expression in video. Image and Vision
     Learning, Indiana, 2006.                                          Computing., in press.
[13] R. Cole, S. van Vuure, B. Pellom, K. Hacioglu, J. Ma,      [28]   D. Marr. Vision. W.H. Freeman and Company, San
     J. Movellan, S. Schwartz, D. Wade-Stein, W. Ward,                 Francisco, 1982.
     and J. Yang. Perceptive animated interfaces: First         [29]   T. Minato, M. Shimada, H. Ishiguro, and S. Itakura.
     steps toward a new paradigm for human computer                    Development of an android robot for studying
     interaction. Proceedings of the IEEE, Special Issue on            human-robot interaction.
     Human Computer Interaction, 91(9):1391–1405, 2003.         [30]   F. Movellan, J. R.and Tanaka, T. C., R. C., and E. M.
[14] K. Dautenhahn. Design issues on interactive                       The rubi project: A progress report. under review.
     environments for children with autism. In Proceedings      [31]   J. R. Movellan. An infomax controller for real time
     International Conference on Disability, Virtual Reality           detection of contingency. In Proceedings of the
     and Associated Technologies (ICDVRAT), pages                      International Conference on Development and
     153–161, Alghero, Sardinia, Italy, Sept. 2000.                    Learning (ICDL05), Osaka, Japan, 2005.
[15] T. F., C. A., and M. J. Socialization between toddlers     [32]   J. R. Movellan, F. Tanaka, B. Fortenberry, and
     and robots at an early childhood education center.                K. Aisaka. The RUBI project: Origins, principles and
     under review.                                                     first steps. In Proceedings of the International
[16] I. R. Fasel. Learning to Detect Objects in Real-Time:             Conference on Development and Learning (ICDL05),
     Probabilistic Generative Approaches. PhD thesis,                  Osaka, Japan, 2005.
     UCSD, June 2006.                                           [33]   J. R. Movellan and J. S. Watson. The development of
[17] I. Fenwick and M. D. Rice. Reliability of continuous              gaze following as a Bayesian systems identification
     measurement copy-testing methods. Journal of                      problem. In Proceedings of the International
     Advertising Research, 1991.                                       Conference on Development and Learning (ICDL02).
[18] T. Fong, I. Nourbakhsh, and K. Dautenhahn. A                      IEEE, 2002.
     Survey of Socially Interactive Robots. Robotics and        [34]   Y. Nagai, M. Asada, and K. Hosoda. Learning for
     Autonomous Systems, 42(3-4):143–166, 2003.                        Joint Attention Helped by Functional Development.
[19] B. Fortenberry, J. Chenu, and J. R. Movellan. Rubi:               Advanced Robotics, 20(10):1165–1181, 2006.
     A robotic platform for real-time social interaction. In    [35]   A. Pentland. Socially Aware Computation and
     Proceedings of the International Conference on                    Communication. IEEE Computer, 38(3):33–40, 2005.
     Development and Learning (ICDL04), The Salk                [36]   J. Peter H. Kahn, B. Friedman, D. R.
     Institute, San Diego, October 20, 2004.                           Perez-Granados, and N. G. Freier. Robotic Pets in the
[20] T. Ishida, Y. Kuroki, and J. Yamaguchi. Development               Lives of Preschool Children. Interaction Studies,
     of Mechanical System for a Small Biped                            7(3):405–436, 2006.
     Entertainment Robot. In Proceedings of the 2003            [37]   R. W. Picard. Affective Computing. The MIT Press,
     IEEE International Workshop on Robot and Human                    1997.
     Interactive Communication, pages 297–302, 2003.            [38]   M. Polsfuss and M. Hess. Liking through
[21] T. Kanda, T. Hirano, D. Eaton, and H. Ishiguro.                   moment-to-moment evaluation: Identifying key selling
     Interactive Robots as Social Partners and Peer Tutors             segments in advertisement. Advances in consumer
     for Children: A Field Trial. Human-Computer                       research, 18:540–544, 1991.
     Interaction, 19(1-2):61–84, 2004.                          [39]   M. J. R. and J. S. Watson. Perception of directional
[22] T. Kanda, H. Ishiguro, T. Ono, M. Imai, and                       attention. In Infant Behavior and Development:
Abstracts of the 6th International Conference on
       Infant Studies, NJ, 1987. Ablex.
[40]   B. Robins, K. Dautenhahn, R. te Boekhorst, and
       A. Billard. Robotic Assistants in Therapy and
       Education of Children with Autism: Can a Small
       Humanoid Robot Help Encourage Social Interaction
       Skills? Universal Access in the Information Society,
       4(2):105–120, 2005.
[41]   B. Scassellati. Foundations for a Theory of Mind for a
       Humanoid Robot. PhD thesis, MIT Department of
       Computer Science and Electrical Engineering, 2001.
[42]   J. Susskind, J. Hershey, and J. Movellan. Exact
       inference in robots using topographical uncertainty
       maps. In Proceedings of the second international
       conference on development and learning (ICDL04),
       The Salk Institute, San Diego, October 20, 2004.
[43]   J. Susskind, G. Littlewort, M. Bartlett, J. Movellan,
       and A. Anderson. Human and computer recognition of
       facial expressions of emotion. Neuropsychologia., in
       press.
[44]   F. Tanaka, B. Fortenberry, K. Aisaka, and J. R.
       Movellan. Developing dance interaction between qrio
       and toddlers in a classroom environment: Plans for
       the first steps: (Best Paper Award). In Proceedings of
       the IEEE International Workshop on Robot and
       Human Interactive Communication (RO-MAN), pages
       223–228, Nashville, USA, 2005.
[45]   F. Tanaka, B. Fortenberry, K. Aisaka, and J. R.
       Movellan. Plans for developing real-time dance
       interaction between qrio and toddlers in a classroom
       environment. In Proceedings of the International
       Conference on Development and Learning (ICDL05),
       Osaka, Japan, 2005.
[46]   F. Tanaka, J. R. Movellan, B. Fortenberry, and
       K. Aisaka. Daily HRI evaluation at a classroom
       environment: Reports from dance interaction
       experiments. In Proceedings of the 2006 Conference on
       Human-Robot Interaction (HRI 06), Salt Lake City,
       2006.
[47]   F. Tanaka and H. Suzuki. Dance Interaction with
       QRIO: A Case Study for Non-boring Interaction by
       Using an Entrainment Ensemble Model. In
       Proceedings of the 2004 IEEE International Workshop
       on Robot and Human Interactive Communication,
       pages 419–424, 2004.
[48]   J. S. Watson. The perception of contingency as a
       determinant of social responsiveness. In E. B.
       Thoman, editor, Origins of the Infant’s Social
       Responsiveness, pages 33–64. LEA, New York, 1979.
[49]   J. S. Watson. Contingency perception in early social
       development. In F. T. M. and F. N. A., editors, Social
       perception in infants, pages 157–176. Ables, New
       Jersey, 1985.
[50]   J. Weng, J. McClelland, A. Pentland, O. Sporns,
       I. Stockman, M. Sur, and E. Thelen. Autonomous
       Mental Development by Robots and Animals. Science,
       291(5504):599–600, 2000.
[51]   D. Wolpert, K. Doya, and M. Kawato. A unifying
       computational framework for motor control and social
       interaction. In F. C and W. DM, editors, The
       Neuroscience of Social Interaction, pages 305–322.
       Oxford University Press, Oxford, UK, 2004.
You can also read